Provably Safe Generative Sampling with Constricting Barrier Functions

arXiv:2602.21429v1 Announce Type: new Abstract: Flow-based generative models, such as diffusion models and flow matching models, have achieved remarkable success in learning complex data distributions. …